ReviseFlow

Inspiration

Panic-studying the night before an exam with 50 pages of notes and no time to make flashcards. We wanted an app that does the hard part for you — just drop your notes and go.

What it does

Upload a PDF, get AI flashcards instantly. Wrong answers re-queue automatically so you drill your weak spots. Every 3 correct answers earns a game credit for a Water Sort puzzle brain break. Sessions and progress persist across reloads so you never lose your place.

How we built it

Layer	Tech
Framework	Next.js 15 (App Router)
AI	Groq API — Llama 3.1 8B Instant
Database	Prisma + SQLite
Client state	localStorage
Animations	Pure CSS keyframes + hand-crafted SVGs

Challenges we ran into

PDF parsing in Turbopack — pdf-parse crashes at module load time; required importing directly from its internal lib/ path via createRequire
API quota burnout — exhausted two Gemini API keys during testing, pivoted to Groq for a genuinely free and faster model
Prisma v7 breaking changes — database URL silently moved out of schema.prisma into a new prisma.config.ts file

Accomplishments that we're proud of

🌊 A fully animated underwater world — sharks, jellyfish, tropical fish, anglerfish, manta ray — built entirely with SVG + CSS, no canvas or WebGL
🤿 A diver mascot with contextual speech bubbles on every page
🎮 A credit-reward loop that makes revision feel like progression, not punishment
💾 Session state that survives tab closes and navigation — students can trust it

What we learned

Reward systems don't need to be complex to change how an app feels to use
Pure CSS animation can handle far more complexity than most developers attempt
PDF text quality is the real bottleneck for AI flashcard quality — clean PDFs produce excellent cards